In 1999, Hershey’s celebrated ERP go-live turned into a Halloween horror story. Rushed configurations and siloed training left the confectioner unable to ship an estimated US $100 million in confirmed orders and shaved 8 percent off its share price overnight. Customers had chocolate on back-order; investors had heartburn. The root cause wasn’t SAP’s code - it was fragmented decision-making during implementation. (FinanSys)
The ERP Paradox
Enterprise suites promise an integrated “single source of truth,” but many implementations turn into siloed units - finance modifies one module, supply chain another, HR a third. Integration, it appears, is more about organisational discipline than a technological feature; even the most robust code base can still break down when teams isolate themselves.
Enter Data Mesh - Autonomy and Adhesive
Zhamak Dehghani’s Data Mesh framework embraces domain autonomy - data as a product owned by the people who know it best - but it also insists on two enterprise-wide binders: self-serve data infrastructure and federated computational governance. Think of them as the “integration bus” that keeps a distributed analytics estate from splintering exactly the way many ERPs have. (ontotext.com)
ERP Pitfalls vs. Data Mesh Risks - and the Governance Antidote
| Classic ERP Failure | Analogous Data Mesh Risk | Federated Governance Antidote |
|---|---|---|
| Over-customised modules create brittle hand-offs | Domains publish idiosyncratic schemas and quality metrics | Universal product contracts: shared SLAs for lineage, freshness, privacy |
| Integration testing left to the end | Data products launched before downstream consumers exist | Shift-left contract tests in CI/CD pipelines |
| Training focuses on module features, not process flow | Teams optimise local analytics, ignore enterprise KPIs | Cross-domain architecture reviews tied to company OKRs |
| One-off data fixes balloon maintenance costs | Duplicate datasets proliferate | Central catalog with reuse incentives - “build once, share everywhere” |
Proof in the Field
-
ING Bank utilised an eight-week Data Mesh proof-of-concept to enable domain teams to build their own chat-journey data products on a governed, self-serve platform, thereby accelerating time-to-market for new insights while maintaining compliance. (Thoughtworks)
-
Intuit surveyed 245 internal data workers and found nearly half their time lost to hunting for owners and definitions in a central lake. Their Mesh initiative reorganised assets into well-described data products, cutting discovery friction and sparking a “network effect” of reuse across thousands of tables. (Medium)
These early adopters report shorter model-validation cycles, lower duplicate-storage spend, and more transparent audit trails - outcomes eerily similar to what successful ERP programs aimed for but rarely achieved.
Four Steps to Build Mesh-Ready Governance
-
Codify the contract. Publish canonical event and entity models (customer, invoice, shipment) with versioning and SLA dashboards visible to every team.
-
Automate policy as code. Inject lineage capture, PII masking, and quality gates into every pipeline - no opt-out, no manual checkpoints.
-
Create integration champions. Rotate enterprise architects or senior analysts into each domain squad to act as diplomats for cross-team reuse.
-
Measure the mesh, not the modules. Track lead time from data request to insight, re-work hours saved, and incident MTTR. Celebrate improvements to the network, not just local deliverables.
Board-Level Takeaway
Domain autonomy without enterprise glue is a recipe for déjà vu - yesterday’s ERP silos reborn in cloud-native form. Treat federated governance as critical infrastructure, fund it like an R&D platform, and hold leaders accountable for both local agility and global coherence.
Call to action: At your next exec meeting, list the three datasets underpinning your highest-stakes AI initiative. If none has (1) a named product owner, (2) a published contract, and (3) automated policy enforcement, your “unified” future is already fragmenting. Invest in the strands before the system snaps.